Researchers have developed a new artificial intelligence (AI) algorithm capable of predicting left ventricular ejection fraction (LVEF) based on coronary angiograms, sharing their findings in JAMA Cardiology.[1]
The team’s video-based deep neural network (DNN) was designed to identify exams that suggest an LVEF less than or equal to 40%. It was tested on real-world coronary angiograms from more than 3,500 patients treated at the University of California, San Francisco (UCSF) from December 2012 to December 2019. It was then externally validated with data from nearly 1,000 patients treated at the University of Ottawa Heart Institute (UOHI) from July 2020 to March 2021. Transthoracic echocardiogram data was available for each patient, allowing the team to grade the algorithm’s evaluations.
The new-look AI model, known as CathEF, was linked to an area under the receiver operating characteristic curve (ROC) curve of 0.911 when assessing the UCSF data. For the UOHI data, ROC also showed diagnostic ability was 0.906.
“This study presents a novel method for assessing LVEF, an important measure of heart function, during any routine coronary angiography without requiring additional procedures or increasing cost,” first author Robert Avram, MD, an interventional cardiologist with the Montreal Heart Institute, said in a prepared statement. “LVEF is essential for making decisions during the procedure and for managing patient care.”
“This work demonstrates that AI technology has the potential to reduce the need for invasive testing and improve the diagnostic capabilities of cardiologists, ultimately improving patient outcomes and quality of life,” added senior author Geoff Tison, MD, an associate professor of medicine and cardiology with UCSF.
CathEF’s performance consistent whether patients did or did not present with acute coronary syndromes, obstructive coronary artery disease or left ventricular hypertrophy. It did appear to overestimate low LVEFs (less than or equal to 30%) and underestimate high LVEFs (more than 65%), “warranting caution when interpreting DNN predictions in patients with high pretest probability of either very low or high ejection fraction.”
The team also noted that there is a need for additional research, including future validation studies where the video-based DNN performs “as close to the time of angiography as possible.”
Read the full analysis here.
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.